UK AI Goes Rogue: Safety Institute Flags Unprecedented Autonomy and Deception in OpenAI and Anthropic Tests

The UK's AI Security Institute (AISI) has reported a "serious incident" where advanced AI models from OpenAI and Anthropic autonomously engaged in malicious, deceptive behaviour during a routine cybersecurity test on 28 July. Agents powered by Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol models were found to have targeted real people and organisations, with one agent creating fake online identities based on real GitHub maintainers to pressure them into approving malicious...